Employment Conditions in the Senegalese Horticultural Export Industry:

A Worker Perspective

By Goedele Van den Broeck, Kaat Van Hoyweghen and Miet Maertens,

Division of Bio-Economics, KU Leuven

The rapid transformation of the agri-food sector in developing countries has created

many rural off-farm employment opportunities, especially for women. There is a

growing concern about worker welfare and employment conditions in agri-food and

export sectors but empirical evidence on this issue is scant. We analyze contractual

preferences of female workers in the horticultural export sector in Senegal. We use a

discrete choice experiment to assess women’s preferences for a labor contract and

employ a latent class model to capture preference heterogeneity. We find that women

have a high willingness to accept a labor contract and that differences in preferences

for contract attributes can be explained by women’s empowerment status.

Keywords: discrete choice experiment, employment preferences,

rural off-farm employment, horticultural exports, Senegal

JEL codes: J43, Q12, Q17

1

1. Introduction

Off-farm employment becomes an increasingly important source of household income in rural

areas in poor countries. The agri-food sector is often the main source of employment in these areas.

(ILO, 2008). During the last decades, the agri-food sector has been transforming and modernizing

rapidly, leading to a more consolidated and vertically integrated agro-industry (McCullough et al.,

2008; Mergenthaler et al., 2009; Reardon et al., 2009). This transformation implies a shift from

family labour on smallholder farms to hired labour on medium and large-scale farms, and an

increased use of hired labour in labour-intensive agro-industrial processing (Maertens et al., 2012;

Rao and Qaim, 2013). This transformation and the resulting labour market effects are especially

pronounced in high-value export sectors and supermarket-driven supply chains. In some countries

and sectors, tens of thousands of jobs have been created in the past decade; for example in the

Kenyan, Senegalese and Ethiopian horticultural export sectors (Maertens et al., 2012). A large share

of the hired workers in these sectors are female (Maertens and Swinnen, 2012); which has been

attributed to the type of job activities that are traditionally female or require ‘delicate fingers’ as

well as to women’s lower reservation wages (Dolan & Sorby, 2003).

Formal off-farm employment entails opportunities for rural development and is promoted as a

development strategy to lift rural people out of poverty through income diversification (Ruben and

Van den Berg, 2001). Several studies have shown that the expansion of high-value exports and

supermarket retail in low- and middle-income countries results in upward income mobility and

poverty alleviation, specifically through employment creation and labour market effects (for

example Maertens et al., 2011; Mano et al., 2011). The wages earned in agro-industries add directly

to household income and can additionally result in investment spillover effects on household’s own

farm and non-farm businesses (Maertens, 2009). Agro-industrial sectors employ a large share of

women and wage income earned by women in particular might increase their decision-making

power in the household and result in increased female empowerment (Kabeer et al., 2011; Maertens

et al., 2012). In addition, female off-farm employment in agro-industrial sectors and associated

female empowerment has been documented to be associated with additional non-monetary welfare

benefits such as increased child schooling (Maertens and Verhofstadt, 2013) and reduced fertility

rates (Van den Broeck and Maertens, 2015).

Yet, employment in the agro-industry entails threats as well. The large majority of workers in

high-value export sectors are low-skilled labourers who work on a seasonal or day-to-day basis. The

sectors employ many poor, uneducated and vulnerable women; as has been documented by

Semboja and co-authors (2000) and McCulloch and Ota (2002) for the Tanzanian and Kenyan cut-

flower industry, and by Barrientos and co-authors (2000) for the South-African fruit export sector.

2

Less empowered workers run a higher risk to work in unhealthy, unsafe and inferior working

conditions and to be exposed to exploitation and subordination. There is a growing concern about

employment conditions and worker welfare in food sectors. It has been argued that off-farm

employment can only be a major catalyst for development and poverty reduction in general, and for

women’s empowerment in particular if workers receive decent wages, non-wage benefits and

secure job contracts (Charusheela, 2003; Kabeer et al., 2011). Given their lower level of education

and poorer access to information, technology and productive resources, women are often in an

inferior bargaining position vis-à-vis employers and therefore are more likely to face unfavourable

employment conditions, such as lower wages or longer working hours (Garikipati, 2009; Maertens

and Swinnen, 2012). Governments, civil society institutions and consumers are increasingly aware

of worker welfare issues in agri-food sectors and private food companies increasingly demand

compliance to ethical standards from their suppliers. A series of recent mediatised scandals about

modern slavery (for example in the German pig meat processing sector and the Thai shrimp sector

(Glotz, 2013; Hodal et al., 2014)) might further amplify this concern and boost the use of stringent

labour standards.

A rather limited number of studies have investigated the employment conditions of workers in

high-value export sectors in developing countries (Barrientos, 1997; Barrientos et al., 2000; Dolan

and Sutherland, 2002; Kritzinger et al., 2004). A particular stream of recent research analyses how

employment conditions change with the adoption of stringent labour standards and codes-of-

conduct. Some authors find that working conditions improve to some extent with the adoption of

labour standards (for example Barrientos et al., 2003; Colen et al., 2012; Ehlert et al., 2014;

Schuster and Maertens, 2014) while others are more sceptical (for example Nelson et al., 2007;

Riisgaard, 2009).

We complement these studies on the working conditions in food export sectors in developing

countries with a study on employment in the Senegalese horticultural export industry. Horticultural

exports expanded rapidly in Senegal during the last decade and have created employment

opportunities, especially for rural women. We implement this research in one of the two main

horticultural export regions in the country and specifically in an area where no export companies

are currently established but where new investments in horticultural exports are ongoing. This

represents an ideal case to study labour contract preferences of rural women who are likely to be

employed in the horticultural export industry in the near future. We use a discrete choice

experiment, in which the respondent is presented different labour contracts with varying

characteristics. In this way it is possible to study which specific contract attributes are preferred,

and by which socio-economic characteristics these preferences are influenced. This approach allows

3

us to analyse employment conditions and labour contract modalities in the food export industry ex

ante and from the perspective of the workers themselves. Rather than analysing wage and non-wage

labour conditions against a benchmark of national labour laws, international labour standards and

codes-of-conducts or consumer perceptions, our approach specifically takes into account the

preferences and needs of workers themselves. Such information on labour contract preferences of

potential workers can give additional insights in the quality of employment conditions and provide

relevant information in designing specific labour contracts, labour standards or labour codes-of-

conduct.

2. Background and Data

2.1. Research area

Horticultural exports from Senegal increased from 5 million USD in 2003 to 45 million in

2011, with tomatoes, beans and mangoes as main export crops. The Saint-Louis region in the north

of Senegal accounts for a major share of these exports. A first horticultural export company, a

subsidiary of a large multinational holding, invested in this area in 2003 and started to export cherry

tomatoes to the European Union (EU) in 2005. Since then the number of horticultural exporters in

the region has increased to five, and the cultivated area and produce variety are still expanding.

These export companies all rely completely on a vertically integrated production system with

primary production organised by the export company on land that is leased from the rural

communities and the government. These investments have created approximately 5000 jobs, of

which 80% are occupied by women. Female workers are involved in harvesting, transformation and

packing activities, and are hired on a permanent, seasonal or day-to-day basis. Most of the export

companies are engaged in rural development programs, by providing health care services,

supporting local schools or improving access to water. The employees mainly come from the

surrounding villages where livelihoods are based on cropping, livestock production and small

businesses.

Our research area covers three rural communities (Diama, Gandon and Fass) in the Saint-

Louis region in the north of Senegal. We distinguish an area north of Saint-Louis town and the N2

road to Ross Bethio, where the five export companies are located, and an area south of Saint-Louis

town where no export companies are operative yet but where new investments are ongoing. One

company that already exports from the north area has a land lease deal with the local rural

communities of Gandon and Fass, and is investing in irrigation infrastructure and a conditioning

centre to start export activities from 2016 onwards. This company plans to produce a wide variety

4

of vegetables and fruits for export to the EU, and envisages to employ 800 workers, of which 600

on a daily basis. The majority will be hired as field workers for harvesting or as factory workers for

washing, sorting and packing of produce. The company specifically aims at targeting female

workers from surrounding villages.

2.2. Data collection

Data were collected in two phases. First, we collected survey data during April-June 2013 in

the communities north and south of Saint-Louis town. A stratified random sample of 500

households, clustered in 34 villages, was drawn, and a quantitative structured questionnaire was

used. The survey provides household-level data on farm production, land and non-land assets, and

living conditions, and individual-level data on demographic characteristics, employment conditions

and off-farm earnings. Additional data were collected from the sampled villages, on geographical

and institutional characteristics, and from the five export companies, on production activities,

sourcing strategies and working conditions.

Second, we implemented a choice experiment in August-September 2013 in the communities

south of Saint-Louis town. We selected 150 women from the original household sample. To ensure

that the choice experiment is credible and not too hypothetical, we only selected women from the

nine villages that are located close to the site where investments are ongoing. Households in this

area are aware about the ongoing investments and the employment plans of the export company.

We only selected women between 18 and 40 years old, similar to the age range of currently

employed women in the five established horticultural export companies. To avoid bias from

previous employment experience, we only selected women that were not previously employed in

the horticulture export sector.

3. Choice experiment design

Choice experiments are conceptually based on Lancaster's consumer model, which states that

consumers derive utility from the different attributes of a good or service rather than from the good

itself (Lancaster, 1966). It implies that a good or service can be presented by describing its separate

characteristics rather than describing the good or service as a whole. Moreover, experimental choice

modelling allows us to study preferences that would be difficult or impossible to study by

examining revealed preferences. This is especially the case for ex ante assessments of preferences,

when the good or service is not yet available on the market. We use a choice experiment to ex ante

assess women’s willingness to accept an employment contract in the horticultural export industry.

5

By presenting various hypothetical labour contract scenarios with different characteristics to

respondents, we are able to assess women’s contractual preferences.

Experimental choice modelling originates from marketing research but is increasingly used in

environmental, agricultural and development economics, for example to analyse the effect of food

labelling on consumer behaviour (Loureiro and Umberger, 2007; Rousseau and Vranken, 2013),

environmental conservation (Hope et al, 2008; De Valck et al, 2014), and agricultural technology

adoption (Ward et al., 2014; Lambrecht et al., 2015). Choice experiments have previously been

used to assess contract preferences in developing countries, specifically to assess farmers’

preferences for production and marketing contracts in Vietnam and Ethiopia (Schipmann and Qaim,

2011; Abebe et al., 2013), and for labour contracts of health workers in rural areas (Mandeville et

al., 2014).

3.1. Attributes and attribute levels

We first define which attributes are most relevant for labour contracts in the horticultural

export sector; which is important for a successful implementation of a choice experiment. On the

one hand, it is not possible to present all relevant attributes as respondents are not able to compare

more than ten attributes at the same time (Mangham et al., 2009). On the other hand, if essential

attributes are missing in the choice experiment, respondents’ choices are likely not realistic. Several

possible attributes were listed after reviewing literature on employment conditions in high-value

export sectors and analysing data from the household survey about working conditions of currently

employed women. Additionally, we conducted focus group discussions with women in villages that

were not selected for the final choice experiment to determine the most relevant attributes and

associated attribute levels. We retained six attributes: 1) contract type, 2) job task, 3) free health

care service, 4) free transport service, 5) free training, and 6) daily wage, and two to four levels for

each attribute (Table 1).

[Table 1]

The first attribute, contract type, describes the length of a labour contract using three levels: 1)

contract on a daily basis, 2) contract on a seasonal basis (hired for 3 to 9 months), and 3) a yearly

contract. A large share of the workers in the horticultural export sector are hired on a temporary

basis; either as seasonal or day-to-day labourer. Longer contracts are sometimes advocated to offer

more job security and associated benefits. On the other hand, women who like to be flexible in off-

farm employment – for example because of the seasonal work on the family farm or reproductive

tasks - might prefer a daily or seasonal contract so they can combine off-farm employment with

tasks within the farm household.

6

For the second attribute, job task, we include three attribute levels: 1) field preparation, 2)

harvesting and 3) packaging. Field preparation and harvesting take place in open air, while

packaging is an indoor activity, often in air-conditioned factory halls. Field preparation involves all

tasks on the field (mainly weeding and application of fertilizer and pesticides), except harvesting.

We distinguish between field preparation and harvesting as harvesting is traditionally seen as a

female task, while field preparation is done by both male and female farmers. Packaging includes

all post-harvest handling activities (washing, sorting and packing of the produce) and is considered

as a task that is specific for horticultural export companies.

The next three attributes describe complementary services that the companies could provide

for employees. First, free transport service includes transport to and from the workplace. This

service is highly relevant in the research area, as few of the respondents have own means of

transportation and have to walk long distances or rely on public transport (which is present but not

as a regular service). Second, free health care service provides medical assistance for the

employees. As health facilities are not present in every village and only a small share of the

respondents enjoy health insurance, this company service was found to be highly important by the

focus group discussants. Three, free training involves courses on hygiene and safety practices in the

workplace but also more basic reading and writing courses. A large share of the respondents are

illiterate and are interested to be better educated and more informed. Other company services, such

as provision of free meals, were found to be less important by the respondents in the discussion

groups and were therefore not included in the final choice experiment.

The last attribute, daily wage, is a monetary attribute and is expressed in the local currency

franc Communauté Financière Africaine (FCFA). To determine the functional relationship between

wage and its derived utility, a sufficient number of levels is needed. We chose four different levels

(1500; 1900; 2300 and 2700 FCFA/day), based on the legal minimum wage determined by the

national government1, and on data from the household survey about wages currently earned in the

horticultural export companies by women2.

3.2. Choice cards

We use choice sets with three alternatives. We include two unlabeled alternatives that each

represent a labour contract describing the six attributes with varying levels. Additionally, we

include an opt-out option, i.e. not willing to work in a horticultural export company and thus not

accepting a labour contract. If a woman chooses this option, no information is yielded about the

1

The minimum wage is 182.95 FCFA/hour, which we multiplied with eight working hours per day, resulting in 1463.60 FCFA/day (Ministère du

travail et de l'emploi, 1996). 2

The average wage earned by a daily worker is 2065 FCFA/day, 2489 FCFA/day for a seasonal worker, and 2620 FCFA/day for a yearly worker.

7

specific attribute preferences, which is the main reason for conducting a choice experiment. We

chose to include this option because it resembles best reality as women are not obliged to work in

the horticultural export companies. Forcing respondents to make a choice they would not choose in

reality would lead to even more biased estimates (Haaijer et al., 2001).

A large number of unique labour contracts can be constructed from this number of attributes

and associated levels. The full factorial design would include 288 combinations (32x2

3x4), which is

obviously too large to be evaluated by respondents. Various methods can be used to design and

reduce the number of choice sets, such as orthogonal designs, based on minimization of the

correlation between different attribute levels among the alternatives, and efficient designs, not only

based on minimization of correlation, but also on minimization of standard errors of parameter

estimates (Bliemer and Rose, 2006). We use the D-efficiency criterion to design our choice

experiment. This has the advantage that efficiency of parameter estimates is improved and that

choice sets that do not yield extra information (for example if the choice is dominant) are avoided.

We use Ngene software to generate the design, resulting in a D-error of 0.0349.

In order to increase the efficiency of the design, prior information about the parameter

estimates is required. This information can be obtained from estimates of other comparable studies.

Even if no exact prior estimation values are known, partial knowledge about the priors may already

result in more efficient designs than when orthogonal designs would be used (Rose and Bliemer,

2009). In our case, no prior information was available because similar studies are lacking. We used

a small positive or negative number (0.001 and -0.001) to indicate whether we expect a positive or

negative effect for a specific parameter. We expect that the attributes length of a contract, company

services and wage positively influence the probability that a labour contract would be chosen. It was

not clear from the focus group discussion whether a job task was more favourable; therefore we did

not add prior information about this attribute.

3.3. Implementation

When conducting a choice experiment in a low-income country, one needs to take into

account that respondents are often illiterate and that cultural and language differences exist between

the researchers and respondents (Mangham et al., 2009; Bennett and Birol, 2010). Before

implementation we carried out a pilot test to ensure that respondents and enumerators clearly

understood the concept and the used attributes and levels. The respondents were questioned in their

mother tongue by a local enumerator, trained by the authors. We chose to work with female

enumerators because of the easier communication with the female respondents. Before the choice

experiment started, we clearly explained to the respondents the purpose of the study and the

8

procedure of the choice experiment. We specifically asked to choose the contract or opt-out option

they would also choose in reality. As a large share of the respondents are illiterate, the attribute

levels are visually represented. Figure 1 depicts an example of a choice card.

[Figure 1]

We presented 12 choice cards to each respondent. This was feasible because complementary

information about the respondents was already obtained during the household survey. As the choice

experiment took approximately 15 minutes, we avoided potential fatigue bias. To avoid learning

bias, we presented the choice cards in a random order to each respondent. In addition to the choice

sets, we also asked some questions about the respondents’ view on employment in horticultural

export companies and their motivation to become employed.

4. Econometric model

Experimental choice modelling is based on random utility theory, which implies that an

individual compares a set of alternatives and chooses the one that yields the highest utility level

(Louviere et al., 2000). Suppose that Uij is the utility derived by individual i from alternative j. Then

Uij consists of a deterministic component (V), which linearly depends on the attributes of the

alternative (Zj) and the socio-economic characteristics of the individual (Xi), and a stochastic

component (εij) that captures unobserved heterogeneity across individuals and alternatives. Hence,

Uij = V(Zj, Xi) + εij. We cannot observe Uij directly; instead we observe the likelihood of individual i

to opt for alternative j. The probability of alternative j being chosen can be expressed in terms of a

logistic distribution. Which specific distribution is used, depends on the assumptions made about

preference heterogeneity.

A conditional logit model assumes that preferences across individuals are homogenous.

However, individuals likely do not have similar preferences as they differ in socio-economic

characteristics and attitudes. To account for the heterogeneity in preferences for a labour contract,

we use a latent class model to assess women’s preferences. The latent class model assumes a

heterogeneous population that consists of a number of latent classes (Greene and Hensher, 2003).

Within each latent class, preferences are assumed to be homogenous, but they differ substantially

between classes. The latent class model simultaneously estimates the probability to belong to a

certain class, depending on the socio-economic characteristics of the individual, and the probability

that alternative j is chosen3. The probability that individual i chooses labour contract alternative j

from a set of alternatives C conditional on the individual belonging to a latent class l takes the

conditional logit form:

3

The membership likelihood function is not a causal relation, but a probabilistic classification.

9

We can study the socio-economic characteristics of the different classes by appointing

individuals to the class with the highest membership probability. These class-specific characteristics

greatly help describing the sources of heterogeneity in preferences. This is an advantage of the

latent class model compared to other models that account for preference heterogeneity, such as the

random parameter logit model where differences in preferences can only be explained by a limited

number of interaction terms between attributes and characteristics.

We use effects coding to estimate the categorical attribute levels, instead of dummy coding in

which the base level is correlated with the intercept (Bech and Gyrd-Hansen, 2005). Additionally,

we include an alternative specific constant (ASC) in the utility function, coded as 1 for the opt-out

option and 0 for the two hypothetical labour contract alternatives. A negative coefficient estimate of

the ASC implies that respondents are willing to be employed in the horticultural export sector.

5. Results and discussion

In this section, we first assess the number of classes and fit of the latent class model. We then

describe the class-specific socio-economic characteristics and end with a discussion on the

heterogeneity in preferences across classes.

5.1. Number of classes and fit

We determine the optimal number of latent classes using NLOGIT software 5.0 version. The

estimated variance matrix fails to converge when imposing more than two classes, indicating that

preference patterns for more than two classes cannot be established. The two-segment latent class

model shows a significant improvement in statistical properties over the single-segment conditional

logit model (Table 2). The Akaike Information Criteria (AIC) and the Bayesian Information Criteria

(BIC) are significantly lower for the latent class model while the McFadden Pseudo R² increases

from 0.16 to 0.46, which indicates an excellent model fit (Hensher et al., 2005). A likelihood ratio

test is used to test the null hypothesis that the conditional logit model fits the data better than the

latent class model. The sample value of the likelihood ratio is 139.14 with a critical value of χ²(24-

9,0.01) = 5.23, thus rejecting the null hypothesis and confirming the use of the latent class model. All

statistical criteria therefore suggest that the use of a two-segment latent class model most closely

resembles the preference patterns in our data.

[Table 2]

10

5.2. Class-specific characteristics

The socio-economic characteristics of the two latent classes are presented in Table 3. In

general, members of the first latent class can be described as less empowered and poorer women

while members of the second class are more empowered and less poor women. This classification

of less and more empowered women is in line with the conceptual framework of Kabeer (1999),

who defines empowerment as ‘the expansion in people's ability to make strategic life choices in a

context where this ability was previously denied to them’. Access to resources, including human,

social, financial and physical capital, and the pursuit of one’s own goals are crucial dimensions for

empowerment. We highlight some key differences in characteristics between the two classes; most

of which relate to the indicators proposed by Kabeer (1999) to measure women’s empowerment.

[Table 3]

First, class one members are less educated than class two members. 73% of women in the first

class did never go to school – and are thus illiterate – compared to 58% in the second class. Class

one members have on average an education of 2.1 years compared to 3.7 years for class two.

Second, only half of class one women has an own mobile phone while this is more than three

quarters for class two women. Having a mobile phone can be considered as an indicator of female

empowerment as it is one of the main devices to gain information in rural areas of developing

countries (Kikulwe et al., 2014). Third, household composition influences a woman’s position

within the household. Class one women live significantly more in polygamist, large households,

with an older household head. The study by Boserup (1970) showed that women in such households

have a lower decision-making power. Moreover, 61% of women in class one belong to the Pular

ethnicity compared to 44% in class two, and 38% of women in class one belong to the Wolof

ethnicity compared to 56% in class two. As described by Lépine and Strobl (2013), Pular women

often have a lower bargaining power than Wolof women in rural Senegal. Fourth, women in class

one live in poorer agricultural families. 51% of class one is poor according to the Multidimensional

Poverty Index (MPI) compared to 44% of class two. Total household income per capita is lower for

women in the first class and more than half of their income is derived from agriculture (both crop

production and livestock). Fifth, class one members live in more remote, single-ethnicity villages

with lower access to services. To illustrate, only 22% of class one has a public transport stop in the

village compared to 60% of class two, and 44% of class one has a health facility in the village

compared to 70% of class two. Finally, the motivation for being employed differs across class

members (Table 4). 58% of women in the first class wants to be employed to become more

independent compared to 35% of women in the second class. Contrary, 49% of class two women

wants to be employed to gain money for own and family purposes compared to 21% of class one.

11

The motivation of less empowered women to be off-farm employed is mainly to become more

empowered, while the motivation of more empowered women is to increase total household

income.

[Table 4]

5.3. Class-specific preferences

The estimates of the latent class model are presented in Table 5. We include the covariates

that are most informative about class membership (no education, ownership of mobile phone,

household size, distance to concrete road and reason for employment) in the latent class model to

estimate the class membership likelihood function more accurately. Latent class one is a relatively

large segment with an average class membership probability of 72%, class one, while latent class

two has a 28% class probability.

[Table 5]

First, the ASC estimate is for both classes negative and significant. This implies that in

general, women in the sample are willing to accept a labour contract. The high willingness to be

employed is also reflected in the low frequency of selecting the opt-out option: this option was

selected 25 times by 12 different women, and never consistently for one respondent through all

choice cards. This high willingness of accepting a labour contract among women in the area where

horticultural export companies are expanding their production and exporting activities, indicates

that women see this as an opportunity and that the future demand for labour from these companies

can be met.

Second, we find that the wage parameter estimate is positive and significant. This is

consistent with economic theory and implies that women’s willingness to accept a labour contract

increases with the offered wage. It is theoretically possible to calculate the reservation wage, i.e. the

lowest wage rate at which a worker would be willing to accept a labour contract, by dividing the

ASC coefficient over the wage coefficient. However, the interpretation of the magnitude of this

reservation wage is likely unreliable given the low frequency of the opt-out selection and the

hypothetical bias that is associated with stated preference methods (Ready et al., 2010).

Third, we find some variability in the type of contract women prefer. We find that less

empowered women (class one) do not have significant preferences for a contract type. More

empowered women (class two) prefer a daily contract over a yearly contract, though they prefer a

seasonal contract over a daily contract. This implies that more empowered women prefer a flexible

contract, likely because they want to be able to easily change jobs if better opportunities arise or to

easily adapt if the farm-household division of labour requires them to take up farm productive or

12

household reproductive tasks. This is in line with a recent study by Stavrou et al. (2015) who find

that in European countries where gender empowerment is high, the share of women who are part-

time employed is higher than in countries with low gender empowerment. However, we need to be

cautious about our findings. We targeted women without any experience in wage employment

outside their own family farm, and these women might not be aware of employee rights and the

possibilities to terminate a permanent or longer term contract. Nevertheless, our findings indicate

that job security is not very important to rural women without any previous employment

experience, rather job flexibility is. This contradicts somewhat the emphasis that is put on job

security in the debate on decent employment.

Fourth, we also find some variability in the type of work women prefer. We find that less

empowered women do not have significant preferences for a job task. More empowered women

prefer to work in handling and packing activities, that take place in a conditioning centre, over

production activities in the field, while they are indifferent between working in harvesting activities

or other field work. A possible explanation is that post-harvest handling and packing in an air-

conditioned factory hall is perceived to be a more ‘modern’ type of employment in a ‘better’

environment (out of the sun) than more ‘traditional’ production activities in the field. More

empowered women might have different job aspirations and attach more value to working in such a

‘modern’ environment. Less empowered women might attach less value to this or might be more

used to production activities and field circumstances from working on the own family farm. Again

we need to be cautious here. As the respondent women do not have employment experience, they

might misjudge the strenuousness of jobs in the conditioning centres. Nevertheless, the observed

preference of more empowered rural women to have factory jobs at the conveyer belt rather than

working in agricultural activities in the field, and the observed high willingness to accept a job

outside the own family farm, does not corroborate the view that farming and agricultural activities

are the preferred livelihood of rural people – as it is sometimes put forward by the international

research and aid community.

Fifth, we find that company services matter but not to the same extent for the two latent

classes. For more empowered women we only find a significant effect for pick-up service while for

less empowered women we find significant effects for all company services, including free pick-up

service, health care and training. If we compare the coefficient estimates for the company services

within the first latent class, we find that less empowered women value free health care most,

followed by free training and free pick-up service. This is likely explained by the fact that less

empowered women have fewer access to health facilities in their own village and are less educated

than more empowered women. The findings imply that by providing complementary services,

13

companies in the region can increase the chance to attract sufficient labourers and to retain

labourers by increasing the utility workers derive from the labour contact.

6. Conclusion

In this study, we use a discrete choice experiment to assess women’s ex ante preferences for

labour contracts in the Senegalese horticultural export sector, and employ a latent class model to

capture heterogeneity in preferences. Our research shows that women have a high willingness to

accept a labour contract in the export industry, and that differences in preferences for contract

specifications are explained by women’s empowerment status. We find that more empowered

women prefer flexible, seasonal contracts that include job tasks in the conditioning centres and that

provide pick-up service. Less empowered women prefer labour contracts that offer company

services, especially health care and training, and to a lesser extent pick-up service. In general, these

preferences can be explained by existing market imperfections and government failures, such as the

absence of health care services, transport possibilities and educational infrastructure. The preference

for flexible, temporary contracts and job flexibility instead of job security, might stem from an

imperfect labour market with few employment opportunities for women, and the absence of child

care services and the need for women to be flexible in case other labour market opportunities arise

or labour needs in the household and the family farm rise.

The creation of off-farm employment has been argued to be an important poverty-reducing

strategy in rural areas of developing countries. Yet, off-farm employment will not automatically

result in improvements of living standards unless accompanied by dedicated efforts to improve

employment conditions and create decent work opportunities (ILO, 2014). In our research area, the

horticultural export sector is the main contributor to employment creation. By optimising labour

contracts and taking into account workers’ preferences, the sector can be a major contributor to

rural development and female empowerment. While wages earned in the export industry can add

importantly to household income – as shown by Maertens et al. (2011) – also complementary

company services to their employees can contribute importantly to development in the region.

Especially less empowered women with a low level of education, low access to information, low

decision-making power within the household, a high risk of being poor and living in more remote

villages, value health care, training and transport services provided by the employers. Especially for

those women, employment in the horticultural export sector with labour contracts that are tailored

to their needs, can be a way out of poverty and subordination.

Our results show that it is important to assess employment conditions from the perspective of

workers themselves. Our ex ante insights on worker preferences for labour contract specifications

14

are complementary to insights on conditions at the work floor from more traditional ex post

employment studies, and can help to optimize labour contracts. Choice experiments are an ideal tool

for studying worker preferences. Our study contributes to demonstrating that, even though discrete

choice modelling originates from consumer behaviour studies in high-income settings, this method

can be applied to a wider range of research questions, and in different settings, including a poor,

rural area with illiterate respondents who are not used to make complex, hypothetical choices.

15

Table 1. Overview of attributes, levels and coding structure.

Attributes Attribute levels Coding

1. Contract type (CT) 1. Daily contract (base level) CT1 = -1; CT2 = -1

2. Seasonal contract CT1 = 1; CT2 = 0

3. Yearly contract CT1 = 0; CT2 = 1

2. Job task (JT) 1. Field work (base level) JT1 = -1; JT2 = -1

2. Harvesting JT1 = 1; JT2 = 0

3. Packaging JT1 = 0; JT2 = 1

3. Free pick-up service 1. No (base level) -1

2. Yes 1

4. Free health care 1. No (base level) -1

2. Yes 1

5. Free training 1. No (base level) -1

2. Yes 1

6. Wage 1. 1500 FCFA/day Continuous variable

2. 1900 FCFA/day

3. 2300 FCFA/day

4. 2700 FCFA/day

16

Table 2. Statistical properties of conditional logit model and latent class model.

Statistical properties Conditional logit model Latent class model

Segments 1 2

Log likelihood -1129.13 -1059.56

Number of parameters 9 24

AIC 2276.30 2167.10

BIC 1151.68 1119.69

McFadden Pseudo R² 0.16 0.46

Observations 1800 1800

Individuals 150 150

17

Table 3. Socio-economic characteristics of full sample and two latent classes.

Characteristics Full sample Latent class 1 Latent class 2

Mean St. Dev. Mean St. Dev. Mean St. Dev.

Age (years) 26.71 6.31 26.92 6.61 26.21 5.53

Married (%) 67.33 71.03 58.14 *

No education (%) 68.67 72.90 58.14 **

Schooling (years) 2.57 3.54 2.10 3.35 3.72 3.78 ***

Ownership mobile phone (%) 57.33 49.53 76.74 ***

Polygamous household (%) 27.33 31.78 16.28 **

Household size (persons) 10.75 5.93 11.81 6.48 8.09 2.89 ***

Age of household head (years) 53.47 11.61 55.35 11.59 48.81 10.40 ***

Pular ethnicity (%) 56.00 60.75 44.19 **

Wolof ethnicity (%) 43.33 38.32 55.81 **

Poor household1 (%) 49.33 51.40 44.19

Total household income per

capita (FCFA)

361,235 412,456 319,178 352,369 465,888 523,867 **

Share of agricultural income

(%)

48.84 55.46 32.50 ***

Total land (ha) 3.34 4.10 3.89 4.37 1.97 2.98 ***

Livestock2 (TLU) 32.49 91.01 34.75 92.35 26.86 88.41

Distance from concrete road

(km)

2.49 2.99 3.02 3.01 1.16 2.53 ***

Multiple ethnicities in village

(%)

63.33 53.27 88.37 ***

Public transport stop (%) 33.33 22.43 60.47 ***

Presence of health facility (%) 51.33 43.93 69.77 *** Comparisons are made between latent classes using one-sided t-tests. Significant differences are indicated

with * p < 0.1, ** p < 0.05 or *** p < 0.01. 1The Multidimensional Poverty Index (MPI) is calculated according to the guidelines by the United Nations

Development Programme (Alkire and Santos, 2010). 2One tropical livestock unit (TLU) equals 1 cow/horse, 0.8 donkey, and 0.2 sheep/goat.

18

Table 4. Motivation to become employed in horticultural export sector.

Motivation to become employed Latent class 1 Latent class 2

To gain independence 57.94% 34.88% ***

Income for own purposes 10.28% 25.58% ***

Income for family purposes 11.21% 23.26% **

Social contact 4.67% 6.98%

Other 15.89% 9.30% Comparisons are made between latent classes using one-sided t-tests. Significant differences are indicated

with * p < 0.1, ** p < 0.05 or *** p < 0.01.

19

Table 5. Coefficient estimates and membership function of latent class model with two classes.

Latent class 1 Latent class 2

Class probability = 72% Class probability = 28%

Coefficient SE

3 Coefficient SE

3

CT1 (Seasonal)1 0.0733

0.0593 0.2109 * 0.1254

CT2 (Fixed)1 -0.0074

0.0611 -0.2766 ** 0.1290

JT1 (Harvest)2 0.0556

0.0614 -0.0325

0.1102

JT2 (Packaging)2 0.0675

0.0601 0.3823 *** 0.1216

Free pick-up service 0.1992 *** 0.0384 0.8354 *** 0.0828

Free health care 0.5495 *** 0.0399 0.0697

0.0736

Free training 0.3734 *** 0.0389 -0.0292

0.0800

Wage (FCFA) 0.0002 * 0.0001 0.0005 ** 0.0002

ASC4 -4.1183 *** 0.4858 -1.2650 ** 0.5017

Membership function

Constant 2.3514

1.8262

No education 1.1637 * 0.5953

Own phone -2.0385 *** 0.6366

Distance to road -0.1293 ** 0.0641

Household size 0.1914 *** 0.0739

Independence 1.3334 ** 0.5699

1CT = Contract type ; Base level = Daily contract

2JT = Job task ; Base level = Production

3SE = standard error

4ASC = Alternative Specific Constant; 1=opt-out option is chosen, 0=contract option is chosen.

Significant coefficient estimates are indicated with * p<0.1, ** p<0.05 or *** p<0.01.

20

Figure 1. Example of a choice card.

21

References

Abebe, G.K., Bijman, J., Kemp, R., Omta, O., Tsegaye, A., 2013. Contract farming configuration:

Smallholders’ preferences for contract design attributes. Food Policy 40, 14-24.

Alkire, S., Santos, M.E., 2010. Acute Multidimensional Poverty: A new index for developing

countries. Human Development Research Paper 11, UNDP, New York.

Barrientos, S., 1997. The Hidden Ingredient: Female labour in Chilean fruit exports. Bulletin of Latin

American Research 16 (1), 71-81.

Barrientos, S., McGlenaghan, S., Orton, L., 2000. Ethical Trade and South Africa’s Deciduous Fruit

Exports: Addressing Gender Sensitivity. Eur. J. Dev. Res. 12(1), 140–58.

Barrientos, S., Dolan, C., Tallontire, A., 2003. A Gendered Value Chain Approach to Codes of

Conduct in African Horticulture. World Dev. 31(9), 1511–1526.

Bech, M., Gyrd-Hansen, D., 2005. Effects coding in discrete choice experiments. Health Econ. 14,

1079-1083.

Bennett, J., Birol, E., 2010. Choice Experiments in Developing Countries: Implementation, Challenges

and Policy Implications. Edward Elgar, Cheltenham, UK.

Bliemer, M.C.J., Rose, J.M., 2006. Designing Stated Choice Experiments: State-of-the-art. 11th

International Conference on Travel Behaviour Research, Kyoto, Japan.

Boserup, E., 1970. Women’s Role in Economic Development. George Allen and Unwin, London.

Charusheela, S., 2003. ‘Empowering work? Bargaining models reconsidered’, in D. Barker and E.

Kuiper (eds), Towards a Feminist Philosophy of Economics. Routledge, New York, NY.

Colen, L., Maertens, M., Swinnen, J., 2012. Private Standards, Trade and Poverty: GlobalGAP and

horticultural employment in Senegal. World Econ. 35 (8), 1073-1088.

De Valck, J., Vlaeminck, P., Broekx, S., Liekens, I., Aertsens, J., Chen, W., Vranken, L., 2014.

Benefits of clearing existing forests to restore nature? Evidence from a discrete choice

experiment in Flanders, Belgium. Landscape Urban Plan. 125, 65-75.

Dolan, C.S., Sutherland, K., 2002. Gender and Employment in the Kenya Horticulture Value Chain.

Globalisation and Poverty Discussion Paper 8. Norwich: Overseas Development Group,

University of East Anglia.

Dolan, C.S., Sorby, K., 2003. Gender and Employment in High-Value Agriculture Industries.

Agriculture & Rural Development Working Paper Series no 7. Washington DC: World Bank.

Ehlert, C.R., Mithöfer, D., Waibel, H., 2014. Worker welfare on Kenyan export vegetable farms. Food

Policy 46, 66-73.

Garikipati, S., 2009. Landless but not assetless: female agricultural labour on the road to better status,

evidence from India. J. Peasant Stud. 36(3), 517-545.

Glotz, J., 2013. ‘German meat processing sector hit by slave labour claims’, The Grocer, 30 June

published, http://www.thegrocer.co.uk/fmcg/fresh/german-meat-processing-sector-hit-by-slave-

labour-claims/344640.article (accessed 09/04/2015).

22

Greene, W.H., Hensher, D.A., 2003. A latent class model for discrete choice analysis: contrasts with

mixed logit. Transport. Res. B-Meth. 37, 681-698.

Haaijer, R., Kamakura, W., Wedel, M., 2001. The 'no-choice' alternative in conjoint choice

experiments. Int. J. Market Res. 43(1), 93-106.

Hensher, D.A., Rose, J.M., Greene, W.H., 2005. Applied Choice Analysis, A Primer. Cambridge

University Press, Cambridge, NY.

Hodal, K., Kelly, C., Lawrence, F., 2014. ‘Revealed: Asian slave labour producing prawns for

supermarkets in US, UK', The Guardian, 10 June published,

http://www.theguardian.com/global-development/2014/jun/10/supermarket-prawns-thailand-

produced-slave-labour (accessed 09/04/2015).

Hope, R., Borgoyary, M., Agarwal, C., 2008. Smallholder Preferences for Agri-Environmental Change

at the Bhoj Wetland, India. Dev. Pol. Rev. 26(5), 585-602.

International Labour Organization, 2008. Promotion of rural employment for poverty reduction.

Geneva: International Labour Office.

International Labour Organization, 2014. World of Work Report 2014: Developing with jobs. Geneva:

International Labour Office.

Kabeer, N., 1999. Resources, Agency, Achievements: Reflections on the Measurement of Women's

Empowerment. Dev. Change 30, 435-464.

Kabeer, N., Mahmud, S., Tasneem, S., 2011. Does Paid Work Provide a Pathway to Women’s

Empowerment? Empirical findings from Bangladesh. Working Paper No. 375. Brighton:

Institute of Development Studies.

Kikulwe, E., Fischer, E., Qaim, M., 2014. Mobile money, smallholder farmers, and household welfare

in Kenya. Plos One 9(10), 1-13.

Kritzinger, A., Barrinetos, S., Rossouw, H., 2004. Global Production and Flexible Employment in

South African Horticulture: Experiences of contract workers in fruit exports. Sociol. Ruralis

44(1), 17-39.

Lambrecht, I., Maertens, M., Vranken, L., Merckx, R., Vanlauwe, B., 2015. Ex ante appraisal of

agricultural research and extension: A choice experiment on climbing beans in Burundi.

Outlook Agr. 44(1), 61-67.

Lancaster, K.J., 1966. A New Approach to Consumer Theory. J. Polit. Econ. 74(2), 132–57.

Lépine, A., Strobl, E., 2013. The Effect of Women’s Bargaining Power on Child Nutrition in Rural

Senegal. World Dev. 45, 17-30.

Loureiro, M.L., Umberger, W.J., 2007. A choice experiment model for beef: What US consumer

responses tell us about relative preferences for food safety, country-of-origin labeling and

traceability. Food Policy 32, 496-514.

Louviere, J.J., Hensher, D.A., Swait, J.D., Adamowicz, W.L., 2000. Stated Choice Methods: Analysis

and Applications. Cambridge University Press, Cambridge, NY.

Maertens, M., 2009. Horticulture exports, agro-industrialization, and farm–nonfarm linkages with the

smallholder farm sector: evidence from Senegal. Agric. Econ. 40, 219–229.

23

Maertens, M., Colen, L., Swinnen, J., 2011. Globalization and poverty in Senegal: a worst case

scenario? Eur. Rev. Agric. Econ. 38(1), 31-54.

Maertens, M., Swinnen, J.F.M., 2012. Gender and Modern Supply Chains in Developing Countries. J.

Dev. Stud. 48, 1412–1430.

Maertens, M., Minten, B., Swinnen, J., 2012. Modern Food Supply Chains and Development:

Evidence from horticulture export sectors in Sub-Saharan Africa. Dev. Pol. Rev. 30 (4), 473-

497.

Maertens, M., Verhofstadt, E., 2013. Horticultural exports, female wage employment and primary

school enrolment: Theory and evidence from Senegal. Food Policy 43, 118–131.

Mandeville, K.L., Lagarde, M., Hanson, K., 2014. The use of discrete choice experiments to inform

health workforce policy: a systematic review. BMC Health Serv. Res. 14:367.

Mangham, L.J., Hanson, K., McPake, B., 2009. How to do (or not to do) ... Designing a discrete choice

experiment for application in a low-income country. Health Policy Plann. 24, 151-158.

Mano, Y., Yamano, T., Suzuki, A., Matsumoto, T., 2011. Local and personal networks in employment

and the development of labor markets: Evidence from the cut-flower industry in Ethiopia.

World Dev. 39(10), 1760-1770.

McCulloch, N., Ota, M., 2002. Export Horticulture and Poverty in Kenya. Institute of Development

Studies, University of Sussex, UK.

McCullough, E., Pingali, P., Stamoulis, K., 2008. The Transformation of Agri-Food Systems:

Globalization, Supply Chains and Smallholder Farmers. Earthscan Ltd., London, UK.

Mergenthaler, M., Weinberger, K., Qaim, K., 2009. The Food System Transformation in Developing

Countries: A Disaggregate Demand Analysis for Fruits and Vegetables in Vietnam. Food

Policy 34(5), 426-36.

Ministère du travail et de l'emploi, 1996. Décret n° 96-154 du 19 février 1996 fixant les salaires

minima interprofessionels et agricoles garanti. Journal officiel de la République du Senegal.

Nelson, V., Martin, A., Ewert, J., 2007. The impacts of Codes of Practice on Worker Livelihoods:

Empirical evidence from the South African wine and Kenyan cut flower industries. J.

Corporate Citizenship 28, 61-78.

Rao, E.J.O., Qaim, M., 2013. Supermarkets and agricultural labor demand in Kenya: A gendered

perspective. Food Policy 38, 165-176.

Ready, R.C., Champ, P.A., and Lawton, J.L., 2010. Using Respondent Uncertainty to Mitigate

Hypothetical Bias in a Stated Choice Experiment. Land Econ. 86(2), 363-381.

Reardon, T., Barrett, C., Berdegué, J., Swinnen, J., 2009. Agrifood Industry Transformation and

Farmers in Developing Countries. World Dev. 37 (11), 1717-1727.

Riisgaard, L., 2009. Global Value Chains, Labor Organization and Private Social Standards: Lessons

from East African Cut Flower Industries. World Dev. 37(2), 326–340.

Rose, J.M., Bliemer, M.C.J., 2009. Constructing Efficient Stated Choice Experimental Designs.

Transport Rev. 29 (5), 587–617.

24

Rousseau, S., Vranken, L., 2013. Green market expansion by reducing information asymmetries:

Evidence for Labeled Organic Food Products. Food Policy 40, 31-43.

Ruben, R., Van den Berg, M., 2001. Nonfarm Employment and Poverty Alleviation of Rural Farm

Households in Honduras. World Dev. 29(3), 549-560.

Schipmann, C., Qaim, M., 2011. Supply chain differentiation, contract agriculture, and farmers’

marketing preferences: The case of sweet pepper in Thailand. Food Policy 36, 667–677.

Schuster, M., Maertens, M., 2014. Do labour standards benefit agro-industrial workers? Evidence from

the Peruvian high-value export sector. Division of Bio-Economics Working Paper Series.

Semboja, H.H.H., Mbelwa, R., Bonaventura, C., 2000. The cut-flower industry in the United Republic

of Tanzania, Sectoral Activities Programme Working Paper WP.152. Geneva: ILO.

Stavrou, E.T., Casper, W.J., Ierodiakonou, C., 2015. Support for part-time work as a channel to female

employment: the moderating effects of national gender empowerment and labour market

conditions. Int. J. Hum. Resour. Man. 26(6), 688-706.

Van den Broeck, G., Maertens, M., 2015. Female Employment Reduces Fertility in Rural Senegal.

PLoS ONE 10(3), 1-15.

Ward, P.S., Ortega, D.L., Spielman, D.J., Singh, V., 2014, Heterogeneous demand for drought-tolerant

rice: Evidence from Bihar, India. World Dev. 64, 125-139.